Matching payments to insurance claims takes 200 hours a month

"Administrative paperwork" is too vague to explain what the AI actually took over. Money moves through a US dental practice along the path below. Pelle's account only sketches the beginning; we reconstructed the full workflow from public sources.

Here is how a dental bill moves through that system:

Most of those "200 hours of administrative work" are spent manually matching payments to insurance claims.

After speaking with hundreds of small-business owners, Pelle was struck by how similar they were. First, most were reluctant operators. Doctors want to treat patients, plumbers want to fix toilets, and nail technicians want to do nails. Paperwork is the last thing they want to touch, but if nobody else can do it, they have to. Second, every small business relies on the same basic tasks: enter data into a system, move data between systems, communicate with customers, and complete transactions.

He offered a specific example: Dr. Ed Zuckerberg, Mark Zuckerberg's father, who opened a dental practice in Dobbs Ferry, New York. After a full day at work, dinner with his family, and putting the children to bed, he would still go downstairs to his home office at 11 p.m. to handle paperwork.

Software has not eliminated this work. It has merely rearranged it. You get a dashboard, a queue, a form, and perhaps a better place to click. But someone still has to click.

Steijn Pelle · a16z News · 2026-06-03

AI changes that. For the first time, software can understand messy context, operate across systems, and finish the job. So the founders asked whether they could build software that lets a small business run itself.

Half of AI agent activity is software engineering. Healthcare accounts for just 1%

If AI is already so capable, why are humans still typing in every payment? Public data offers an answer. In February 2026, Anthropic analyzed nearly 1 million agent tool calls on its API and classified them by industry.

Half of all agent activity is software engineering. The other half is spread across 16 industries, none above 9%. Healthcare accounts for 1.0%; legal work, 0.9%.

Those small numbers are easy to misread. They point to markets that have barely been tapped. Healthcare AI is still in its infancy.

Anthropic offered a more practical explanation in a footnote to the study. Software moved first because code is easy to verify: run it, and you can see whether it works. Errors are also easier to detect. In law, healthcare, and finance, checking whether an agent performed correctly often requires as much professional expertise as doing the work itself. Trust therefore takes much longer to build.

Dental practices rely on dozens of systems that fail without warning

If the market is so open, why has nobody solved it? Pelle identifies three barriers. The second is the most concrete.

Each works differently—and fails without warning.

Steijn Pelle · a16z News · 2026-06-03

The founders had relevant experience. At Robinhood, Pelle's team worked obsessively to eliminate nearly all manual review from account opening, becoming one of the first brokerages to do it. His co-founder, Frédéric Renken, was the first product hire at Superhuman, where the team scrutinized every point of friction in a new user's first experience.

Their unglamorous playbook: do the job, then install the product in person

How did they make it work? The answer has little to do with models. They embraced two labor-intensive tactics Silicon Valley had long dismissed. Both can be traced to essays Paul Graham wrote more than a decade ago.

Lassie's two co-founders. The stack of papers on the windowsill and the forms spread across the desk are the materials they handled every day for months. Image: a16z News.

700 practices, 49 states, and $10 million in annualized revenue

What did those two unglamorous methods produce? As of June 3, 2026, the company reported these figures.

The last two figures cross-check: 250,000 hours ÷ 700 practices ÷ 12 months ≈ 29.8 hours per practice per month, matching the stated average of 30 hours. Growth has come mainly from word of mouth and customer referrals, adding dozens of practices each month. One customer told them, "You saved my marriage."

The $35 million Series A was led by a16z's Alex Rampell and Olivia Moore. The announcement puts the market at roughly 160,000 dental practices in the US. Pelle uses a broader figure of 500,000 medical practices in his account.

Today, the product automates two steps in the payment flow

Lassie's product page lists three functions. Their position in the earlier payment flow is straightforward.

a16z describes it the same way. Dental is the first market because the pain is acute, the workflow is structured and repetitive, and the dollar amounts are meaningful. The larger ambition extends beyond dentistry, but that comes later.

Two nearby figures can look like the same claim

Pelle says the system aims to replace "$200,000 a year in administrative labor." The actual result reported in the same article is an average of 30 hours recovered per practice each month.

Placed side by side, those figures can make it sound as though the full $200,000 has already been saved. At typical US wage rates for dental administrative staff, 30 hours a month represents several hundred to roughly $1,000 in monthly labor costs—or several thousand to a little over $10,000 a year. $200,000 is the ceiling for the whole category of work. 30 hours is how far the product has reached today. There is still a large gap between them.

One more calculation: $10 million in annualized revenue ÷ 700 practices ≈ $14,300 per practice per year, or roughly $1,190 per month. This is our back-of-the-envelope estimate from two public figures, not an official price. Lassie has not published pricing, and actual rates may vary by practice size.

Two issues go unmentioned throughout Pelle's account: who is liable when the AI files an insurance claim incorrectly, and how the system meets HIPAA requirements for patient data. For a product that directly handles medical billing and patient information, these are the two most important unanswered questions. The article says only that the work must be done correctly, and we found no public answers.

What customers pay for is not quite what the founder pitches

Behind the headline numbers, the real reason customers pay is slightly different. You can see it only by comparing Pelle's account with the investment announcement published the same day. Each identifies a different primary benefit.

Saving time is a feeling. Recovering lost money appears on the ledger. The first makes owners happy; the second gives them a measurable reason to renew every month.

The same announcement draws a hard line around the product: Lassie is not a copilot. A copilot would create more work for practice staff to review. Lassie's goal is to finish the job directly.

Five tactics from this case are directly reusable

Set Lassie itself aside. What remains useful about its approach? Each of the five lessons below follows from the evidence above—not from a vague claim that "vertical AI has potential."

The saved time did not turn into layoffs

One obvious question remains: what happens to administrative staff after dozens of hours of work disappear?

Pelle invokes an old economic idea: the lump of labour fallacy. It describes the mistaken belief that the total amount of work in the world is fixed, so every additional task performed by a machine must remove one from a person.

After observing 700 practices, he says he is convinced this is a fallacy. Owners used the time to expand their practices, improve the patient experience, and do work that had always been more valuable. That is the company's account, unsupported by independent data, though it points in the same direction as another public study.

Lassie says its next two jobs are spreading the message faster—millions of owners still do not know this is possible—and making installation simple enough for any owner to complete alone. After that comes going deeper: doing more work and connecting to more of the tools a business already uses.

Pelle's account ends with a hiring pitch.

A note on the figures: the 700-practice count cited here is from June 3, 2026. At publication, Lassie's homepage said "2,500+ doctors," while its company page still said "700+ practices." Doctors and practices are different units—a practice typically has several doctors—so the two figures cannot be compared directly.

AI can already do sophisticated work, so why are people still entering these payments manually? In February 2026, Anthropic analyzed nearly 1 million agent calls on its API—AI programs that can use tools and finish tasks—and classified them by industry.

Healthcare's 1.0% is easy to misread. It marks a market that has barely opened.

The hard part was not the model. A dental practice connects to dozens of insurer portals, several claims clearinghouses, one practice management system, and one bank account. In Steijn Pelle's words, each operates in its own way and breaks on its own schedule. Their answer was two forms of manual work Silicon Valley had dismissed for years.

Over those months, the founders reconciled millions of dollars in insurance payments, submitted thousands of insurance claims, and billed hundreds of patients. That is how they learned exactly which actions consumed those 200 hours. They personally installed the first 100 customers from Florida to Kansas to Oregon. Only then did they understand how customer 101 could connect the system alone. The two ideas come from Paul Graham's schlep—tedious work nobody wants to do—and Stripe's early on-site installations.

As of June 3, 2026, those two methods had produced the following results:

The last two figures agree: 250,000 hours ÷ 700 practices ÷ 12 months ≈ 29.8 hours per practice per month. All figures come from Lassie and a16z and have not been independently verified.

The article's most valuable insight appears only when the founder's account is compared with the investment announcement published the same day: each side identifies a different primary benefit.

Saving time is a feeling. Recovering lost money is measurable—and gives customers a reason to renew each month. The same announcement draws a hard line around the product: Lassie is not a copilot, an assistant that offers advice and waits for a human decision. A copilot would create more work for practice staff to review.